aboutsummaryrefslogtreecommitdiff
path: root/driver/gator_trace_sched.c
blob: 700dd2ecdf089911be8213851a0f8583f656937f (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
/**
 * Copyright (C) ARM Limited 2010-2015. All rights reserved.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 *
 */

#include <trace/events/sched.h>
#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
#include <trace/events/task.h>
#endif

#include "gator.h"

#define TASK_MAP_ENTRIES		1024	/* must be power of 2 */
#define TASK_MAX_COLLISIONS		2

enum {
	STATE_WAIT_ON_OTHER = 0,
	STATE_CONTENTION,
	STATE_WAIT_ON_IO,
	CPU_WAIT_TOTAL
};

static DEFINE_PER_CPU(uint64_t *, taskname_keys);
static DEFINE_PER_CPU(int, collecting);

/* this array is never read as the cpu wait charts are derived
 * counters the files are needed, nonetheless, to show that these
 * counters are available
 */
static ulong cpu_wait_enabled[CPU_WAIT_TOTAL];
static ulong sched_cpu_key[CPU_WAIT_TOTAL];

static int sched_trace_create_files(struct super_block *sb, struct dentry *root)
{
	struct dentry *dir;

	/* CPU Wait - Contention */
	dir = gatorfs_mkdir(sb, root, "Linux_cpu_wait_contention");
	if (!dir)
		return -1;
	gatorfs_create_ulong(sb, dir, "enabled", &cpu_wait_enabled[STATE_CONTENTION]);
	gatorfs_create_ro_ulong(sb, dir, "key", &sched_cpu_key[STATE_CONTENTION]);

	/* CPU Wait - I/O */
	dir = gatorfs_mkdir(sb, root, "Linux_cpu_wait_io");
	if (!dir)
		return -1;
	gatorfs_create_ulong(sb, dir, "enabled", &cpu_wait_enabled[STATE_WAIT_ON_IO]);
	gatorfs_create_ro_ulong(sb, dir, "key", &sched_cpu_key[STATE_WAIT_ON_IO]);

	return 0;
}

static void emit_pid_name(const char *comm, struct task_struct *task)
{
	bool found = false;
	char taskcomm[TASK_COMM_LEN + 3];
	unsigned long x, cpu = get_physical_cpu();
	uint64_t *keys = &(per_cpu(taskname_keys, cpu)[(task->pid & 0xFF) * TASK_MAX_COLLISIONS]);
	uint64_t value;

	value = gator_chksum_crc32(comm);
	value = (value << 32) | (uint32_t)task->pid;

	/* determine if the thread name was emitted already */
	for (x = 0; x < TASK_MAX_COLLISIONS; x++) {
		if (keys[x] == value) {
			found = true;
			break;
		}
	}

	if (!found) {
		/* shift values, new value always in front */
		uint64_t oldv, newv = value;

		for (x = 0; x < TASK_MAX_COLLISIONS; x++) {
			oldv = keys[x];
			keys[x] = newv;
			newv = oldv;
		}

		/* emit pid names, cannot use get_task_comm, as it's not exported on all kernel versions */
		if (strlcpy(taskcomm, comm, TASK_COMM_LEN) == TASK_COMM_LEN - 1) {
			/* append ellipses if comm has length of TASK_COMM_LEN - 1 */
			strcat(taskcomm, "...");
		}

		marshal_thread_name(task->pid, taskcomm);
	}
}

static void collect_counters(u64 time, struct task_struct *task, bool sched_switch)
{
	int *buffer, len, cpu = get_physical_cpu();
	long long *buffer64;
	struct gator_interface *gi;

	if (marshal_event_header(time)) {
		list_for_each_entry(gi, &gator_events, list) {
			if (gi->read) {
				len = gi->read(&buffer, sched_switch);
				if (len < 0)
					pr_err("gator: read failed for %s\n", gi->name);
				marshal_event(len, buffer);
			} else if (gi->read64) {
				len = gi->read64(&buffer64, sched_switch);
				if (len < 0)
					pr_err("gator: read64 failed for %s\n", gi->name);
				marshal_event64(len, buffer64);
			}
			if (gi->read_proc && task != NULL) {
				len = gi->read_proc(&buffer64, task);
				if (len < 0)
					pr_err("gator: read_proc failed for %s\n", gi->name);
				marshal_event64(len, buffer64);
			}
		}
		if (cpu == 0)
			gator_emit_perf_time(time);
		/* Only check after writing all counters so that time and corresponding counters appear in the same frame */
		buffer_check(cpu, BLOCK_COUNTER_BUF, time);

		/* Commit buffers on timeout */
		if (gator_live_rate > 0 && time >= per_cpu(gator_buffer_commit_time, cpu)) {
			static const int buftypes[] = { NAME_BUF, COUNTER_BUF, BLOCK_COUNTER_BUF, SCHED_TRACE_BUF, ACTIVITY_BUF };
			int i;

			for (i = 0; i < ARRAY_SIZE(buftypes); ++i)
				gator_commit_buffer(cpu, buftypes[i], time);

			/* spinlocks are noops on uniprocessor machines and mutexes do
			 * not work in sched_switch context in RT-Preempt full, so
			 * disable proactive flushing of the annotate frame on
			 * uniprocessor machines.
			 */
#ifdef CONFIG_SMP
			/* Try to preemptively flush the annotate buffer to reduce the chance of the buffer being full */
			if (on_primary_core() && spin_trylock(&annotate_lock)) {
				gator_commit_buffer(0, ANNOTATE_BUF, time);
				spin_unlock(&annotate_lock);
			}
#endif
		}
	}
}

/* special case used during a suspend of the system */
static void trace_sched_insert_idle(void)
{
	marshal_sched_trace_switch(0, 0);
}

static void gator_trace_emit_link(struct task_struct *p)
{
	int cookie;
	int cpu = get_physical_cpu();

	cookie = get_exec_cookie(cpu, p);
	emit_pid_name(p->comm, p);

	marshal_link(cookie, p->tgid, p->pid);
}

GATOR_DEFINE_PROBE(sched_process_fork, TP_PROTO(struct task_struct *parent, struct task_struct *child))
{
	gator_trace_emit_link(child);
}

#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
GATOR_DEFINE_PROBE(sched_process_exec, TP_PROTO(struct task_struct *p, pid_t old_pid, struct linux_binprm *bprm))
{
	gator_trace_emit_link(p);
}

#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 15, 0)
GATOR_DEFINE_PROBE(task_rename, TP_PROTO(struct task_struct *task, char *comm))
#else
GATOR_DEFINE_PROBE(task_rename, TP_PROTO(struct task_struct *task, const char *comm))
#endif
{
	emit_pid_name(comm, task);
}
#endif

#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 35)
GATOR_DEFINE_PROBE(sched_switch, TP_PROTO(struct rq *rq, struct task_struct *prev, struct task_struct *next))
#else
GATOR_DEFINE_PROBE(sched_switch, TP_PROTO(struct task_struct *prev, struct task_struct *next))
#endif
{
	int state;
	int cpu = get_physical_cpu();

	per_cpu(in_scheduler_context, cpu) = true;

	/* do as much work as possible before disabling interrupts */
	if (prev->state == TASK_RUNNING)
		state = STATE_CONTENTION;
	else if (prev->in_iowait)
		state = STATE_WAIT_ON_IO;
	else
		state = STATE_WAIT_ON_OTHER;

	per_cpu(collecting, cpu) = 1;
	collect_counters(gator_get_time(), prev, true);
	per_cpu(collecting, cpu) = 0;

#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 4, 0)
	gator_trace_emit_link(next);
#endif
	marshal_sched_trace_switch(next->pid, state);

	per_cpu(in_scheduler_context, cpu) = false;
}

GATOR_DEFINE_PROBE(sched_process_free, TP_PROTO(struct task_struct *p))
{
	marshal_sched_trace_exit(p->tgid, p->pid);
}

static void do_nothing(void *info)
{
	/* Intentionally do nothing */
	(void)info;
}

static int register_scheduler_tracepoints(void)
{
	/* register tracepoints */
	if (GATOR_REGISTER_TRACE(sched_process_fork))
		goto fail_sched_process_fork;
#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
	if (GATOR_REGISTER_TRACE(sched_process_exec))
		goto fail_sched_process_exec;
	if (GATOR_REGISTER_TRACE(task_rename))
		goto fail_task_rename;
#endif
	if (GATOR_REGISTER_TRACE(sched_switch))
		goto fail_sched_switch;
	if (GATOR_REGISTER_TRACE(sched_process_free))
		goto fail_sched_process_free;
	pr_debug("gator: registered tracepoints\n");

	/* Now that the scheduler tracepoint is registered, force a context
	 * switch on all cpus to capture what is currently running.
	 */
	on_each_cpu(do_nothing, NULL, 0);

	return 0;

	/* unregister tracepoints on error */
fail_sched_process_free:
	GATOR_UNREGISTER_TRACE(sched_switch);
fail_sched_switch:
#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
	GATOR_UNREGISTER_TRACE(task_rename);
fail_task_rename:
	GATOR_UNREGISTER_TRACE(sched_process_exec);
fail_sched_process_exec:
#endif
	GATOR_UNREGISTER_TRACE(sched_process_fork);
fail_sched_process_fork:
	pr_err("gator: tracepoints failed to activate, please verify that tracepoints are enabled in the linux kernel\n");

	return -1;
}

static void unregister_scheduler_tracepoints(void)
{
	GATOR_UNREGISTER_TRACE(sched_process_fork);
#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 4, 0)
	GATOR_UNREGISTER_TRACE(sched_process_exec);
	GATOR_UNREGISTER_TRACE(task_rename);
#endif
	GATOR_UNREGISTER_TRACE(sched_switch);
	GATOR_UNREGISTER_TRACE(sched_process_free);
	pr_debug("gator: unregistered tracepoints\n");
}

static void gator_trace_sched_stop(void)
{
	int cpu;

	unregister_scheduler_tracepoints();

	for_each_present_cpu(cpu) {
		kfree(per_cpu(taskname_keys, cpu));
	}
}

static int gator_trace_sched_start(void)
{
	int cpu, size;
	int ret;

	for_each_present_cpu(cpu) {
		size = TASK_MAP_ENTRIES * TASK_MAX_COLLISIONS * sizeof(uint64_t);
		per_cpu(taskname_keys, cpu) = kmalloc(size, GFP_KERNEL);
		if (!per_cpu(taskname_keys, cpu))
			return -1;
		memset(per_cpu(taskname_keys, cpu), 0, size);
	}

	ret = register_scheduler_tracepoints();

	return ret;
}

static void gator_trace_sched_offline(void)
{
	trace_sched_insert_idle();
}

static void gator_trace_sched_init(void)
{
	int i;

	for (i = 0; i < CPU_WAIT_TOTAL; i++) {
		cpu_wait_enabled[i] = 0;
		sched_cpu_key[i] = gator_events_get_key();
	}
}